AI VISIBILITY
Getting AI Mentions
A buyer asks ChatGPT who to use, and it reads a few names off a list before it reaches yours. That list is your wake: the trail your brand leaves across everyone else’s pages, which the engine reads instead of your hull. You can’t write your own entry on it, and you can’t ask to be called out. You can only widen the wake until the engine has no honest way to skip your name. This page is how a mention is actually decided, why the same brand gets named on one engine and quietly waved through unnamed on another, and the work that moves it.
89.8%
of brands tested earned zero AI mentions across eight engines (Victorious, Q1 2026)
The short version
What an AI mention really is
An AI mention is the moment an engine like ChatGPT, Perplexity, or Gemini names your brand inside its answer. You don’t request one and you can’t buy a guaranteed one. The engine reaches it like a verdict, from the pattern of what independent sources already say about you, so the only durable route is to become the brand those sources keep naming.
This guide is part of our AI visibility hub. It’s written for practitioners, so it skips the intro-level scaffolding and gets to what actually moves a name into an answer.
AI mention statistics 2026
The silence most of your category is sitting in, and why what others publish about you outweighs what you publish about yourself.
~62%
of AI citations never name the brand
The source is linked, the name stays silent. Semrush with Kevin Indig, June 2026
Think of an AI engine as a doorman reading a guest list. When a buyer asks who to trust in your category, it does not invent an answer. It calls a few names off a list that other people wrote, and that list was settled long before this particular buyer walked up. Your job is not to slip the doorman a tip. It is to get written onto the list so often, by enough credible hands, that leaving you off would look like an error.
How a mention is actually decided
Two systems hold that list, and they fill it in different ways. The first is the model’s trained memory, the associations baked in while it read the open web. A brand named again and again across authoritative, independent text gets stored as a strong answer for its category. This layer is slow to build and just as slow to fade. The second is live retrieval, where engines like Perplexity and Google AI Overviews fetch fresh pages at the moment of the question. This layer is fast, and it rewards content that is current, well-structured, and easy to lift.
Here is the part that surprises people. Getting onto the list and getting your name read out loud are two separate things, and the engines disagree on the second. In a June 2026 study by Semrush with Kevin Indig and Growth Memo, ChatGPT cited a source 87% of the time but spoke the brand’s actual name in only about a fifth of its answers. Gemini did the reverse, naming brands in most answers while linking back far less often. So the same standing can render as a glowing unlinked mention on one engine and a silent citation on another. You are not optimizing for “a mention.” You are earning a reputation that each engine then chooses how to voice.
Why you can’t shortcut it
You can’t prompt your way onto the list, because a prompt acts at the moment of the question, while your standing was fixed earlier, in training and in the retrieval index. You can’t keyword-stuff your way on either, because a retrieval engine weighs whether a source looks credible, not how many times a phrase repeats. And you can’t buy your way on with cheap links, because manipulation patterns read as exactly the noise the engine is built to discount. The signal it is hunting for is independent agreement, and that has to be real.
The moves that earn mentions
Five plays do most of the work. Each has a clear reason it earns a place on the list.
1. Get named on the sources the engines already read
Why it works: when a publication that already feeds answers in your niche names you, both layers gain at once. The mention seeps into the trained memory over time and can be retrieved live today. Find which domains keep getting cited for your category, then earn genuine coverage there through real editorial placements, not a syndication dump. The brands the engines name are the ones already written about across the web, not the ones that link to themselves the most.
2. Publish original data worth citing
Why it works: engines reach for specific numbers from named sources, and original research is the one asset only you can own. A single study or proprietary data point that others quote turns you into the source behind the answer instead of one more candidate hoping to be picked.
3. Structure pages so an engine can lift the answer
Why it works: a retrieval engine quotes the passage that most cleanly answers the question. A direct 40-to-60-word answer under a plain question heading, backed by a real figure, is far easier to lift than the same point buried in paragraph nineteen. This is the half of the work you fully control, and it is the half most brands neglect.
4. Use press distribution only for genuine news
Why it works, and where it stops: a release about real news adds a credible, dated signal that engines can pick up. But across the millions of citations studied, press releases make up only around 1% of all AI citations, so distribution is one corroboration input and never the program. A wire about nothing adds noise, and the engines have learned to ignore noise.
5. Keep showing up
Why it works: one mention on one source reads as a coincidence to a model. Repeated, independent mentions across credible sources read as consensus, and consensus is the thing the engine is trying to surface. This is why mention work is a program, not a launch.
A simple sequence to start
If you’re starting cold, run the five plays in order rather than all at once. First, baseline where you stand and find the sources that already feed answers in your niche. Second, make your own pages extractable and well-sourced, because that part is fast and fully yours. Third, publish one piece of original data worth quoting. Fourth, earn coverage on two or three of those key sources. Fifth, set a steady cadence and hold it. The brands that win are not the ones who do everything in week one. They are the ones still on the list in month six.
Where press distribution stops
Press distribution earns a place in the program, but a small one. A release about real news adds a credible, dated signal an engine can pick up. But across the millions of citations studied, press releases make up only around 1% of all AI citations. Treat it as one corroboration input, never the whole effort. A wire about nothing just adds noise the engines have learned to ignore.
How to know it's working
Set a baseline before you change anything. List the questions a buyer in your category would actually ask, run each across ChatGPT, Perplexity, Gemini, and Claude a handful of times, and record two things: who gets named, and which sources get cited. That is your before picture. Re-run it monthly. The number that matters is your share of mentions across that prompt set trending up over a quarter, not one lucky answer on a good afternoon.
Watch the named layer and the cited layer separately, because they move at different speeds. Retrieval mentions can show up within days of publishing fresh, well-structured, well-sourced content, so treat those as your early signal. Trained-memory mentions move on the order of months, since they ride on model updates whose timing nobody outside the labs knows. If your cited count is climbing while your named count lags, that is the normal order of events, not a failure. Keep widening coverage and let the slow layer catch up. For the tooling side of this, our AI visibility tools guide covers what to track it with.
The one number that matters
Your share of mentions across a fixed prompt set, trending up over a quarter. Not one lucky answer on a good afternoon.
What stalls a mention program
Three things quietly kill progress. Blocked crawlers come first: if GPTBot, PerplexityBot, or ClaudeBot can’t fetch your pages, none of the rest counts, so check your logs and any edge rules before anything else. Thin or unsourced content comes next: a page that makes claims without backing them gives a retrieval engine nothing safe to quote. And impatience finishes the job: brands that build coverage for two months, see only retrieval-layer movement, and quit have walked away the week before the slow layer would have turned. Anyone promising you a guaranteed name in the answer by Friday is reading you a forecast they can’t keep.
Related resources in this silo
- LLM Brand Visibility: Own Your AI Narrative
- Best AI Visibility Tools 2026
- Do Press Releases Drive AI Citations?
Part of the AI visibility hub, itself part of the AI SEO guide. When you want this run for you rather than in-house, see how Puffer builds AI citations.
Mentions follow presence. Presence is the work.
If the engines keep reading someone else’s name for your category, the fix isn’t a clever prompt or a one-off wire. It’s steady, credible presence on the sources the models already trust. We at The Puffer build that presence through editorial placements on real publications, contextual backlinks that strengthen your entity, and genuine-news distribution where it earns its keep, all measured against a baseline you can verify yourself. Reach out and we’ll map what earning your mentions would take.
The engines read the wake, not the ship. Widen yours, and the name follows.
Tell us the questions your buyers ask
Tell us the questions your buyers actually ask an engine. We’ll map where your name surfaces today and what it takes to widen the wake.
Frequently asked questions
What is an AI mention?
It’s when an AI engine names your brand inside its answer to a buyer’s question. Mentions come from two systems: the model’s trained associations and its live retrieval of cited pages. Earning them means being genuinely present across the independent sources both systems trust, so the engine reaches your name on its own.
How do I get my brand mentioned by ChatGPT?
Earn independent coverage on the sources ChatGPT’s training and retrieval already lean on, publish original data others quote, and structure your pages so a clean answer is easy to lift. There is no request button. You make yourself the obvious name to read off the list.
Can I pay to be mentioned by AI engines?
Not directly, and you should distrust anyone selling a guaranteed mention. You can invest in the coverage and content that make a mention likely, but the engine decides, and nobody outside the model makers knows the exact weighting between trained memory and live retrieval.
How long does it take to earn AI mentions?
Retrieval-layer mentions can appear within days of publishing fresh, well-sourced content. Trained-memory mentions usually take months, because they depend on model updates. Plan on a three-to-six-month horizon for the slow layer to move.
Do press releases get my brand mentioned by AI?
They contribute, but they don’t carry a program. Press releases account for only around 1% of all AI citations, so a release about genuine news adds a credible, dated signal, while the bulk of the work is steady earned coverage on the sources engines already read.